1. Climate Predictability Tool (CPT) Ousmane Ndiaye and Simon J. Mason cpt@iri.columbia.edu International Research Institute for Climate and Society The Earth Institute of Columbia University

2. The Climate Predictability Tool (CPT) provides a Windows package for :  seasonal climate forecasting  model validation  actual forecasts given updated data Uses ASCII input files Options :  Principal Components Regression (PCR)  Canonical Correlation Analysis (CCA)  Multi Linear Regression (MLR) Help Pages on a range of topics in HTML format Options to save outputs in ASCII format and graphics as JPEG Program source code is now available for those using other systems (e.g., UNIX) OVERVIEW

3. What is CPT? CPT is designed to produce statistical forecasts of seasonal climate using either the output from a GCM, or fields of sea- surface temperatures. Features: Model training Validation Verification Flexible forecasts

4. CPT into context Some of the known CPT courses to date:

5. Why CPT?

6. Choose the analysis to perform: PCR, CCA or MLR SELECTING THE ANALYSIS

7. Both analysis methods require two datasets: “X variables” or “X Predictors” dataset; “Y variables” or “Y Predictands” dataset. IMPUT DATASETS

8. 1. STATION files: This file-type contains : Station_name (without spaces;  16 characters) Latitude (in signed degrees) Longitude (signed degrees) Year (in the first column) Data (missing values should be filled with the same value, -9999 for example) Keywords: STN, LAT, LONG CPT INPUT FILE FORMATS

9. 2. UNREFERENCED or Indices files: The data are not referenced (no latitudes and longitudes): Index_name (without spaces;  16 characters) Year (in the first column) Data (with missing data) Keywords: NAME or YEAR CPT INPUT FILE FORMATS

10. The input files could be easily made using a spreadsheet such as Excel CPT INPUT FILE FORMATS

11. In Excel the file should be saved as: “Text, tab delimited” CPT INPUT FILE FORMATS

12. To select input files just click on browse. SELECTING INPUT FILES

13. CPT opens a browser, which by default looks for data in: C:\Documents and settings\user\Application Data\CPT\DATA\ or the directory specified during installation. You can search for data from any other directory. SELECTING INPUT FILES

14. For gridded and station datasets, CPT lets you choose the spatial domain over which you want to perform your EOF or CCA analysis. In general the domain is known in advance through experience. SELECTING INPUT FILES

15. You proceed in the same way to select your file containing the Y variables (predictands). SELECTING INPUT FILES

16. By default CPT usually starts the analysis from the first years in the X and Y files; note that these years could be different. You would normally set them equal to the latest year in the two files. You should make sure the lag is correct if you cross the calendar year while using the DJF or JFM season, for example. In this case the starting year for file X may need to be one year earlier than for file Y. SETTING THE TRAINING PERIOD

17. You have to specify the length of the training period as well as the length of the cross-validation window. SETTING THE TRAINING PERIOD

18. You have to choose the number of EOFs for the predictor fields used to fit the model. If you set the minimum to be less than the maximum, CPT will find the optimum number of modes between the two numbers. However, if you set the minimum equal to the maximum, then CPT will use that number of modes. SETTING ANALYSIS OPTIONS

19. If you have missing values in your dataset, you need to specify what you want CPT to do with them. MISSING VALUES

20. Next to the Missing value flag box, you need to specify the number in your dataset that represents a missing value. You can choose the maximum % of missing values. If a station has more than that percentage of missing values, CPT will not use that station in its model. You can also choose which method you want CPT to use to replace the values. MISSING VALUES

21. Once you have selected the input files and your settings it is a good idea to save these settings in a project file to recall them later: File => Save By default, CPT saves all the project files in the subdirectory C:\Documents and settings\user\Application Data\CPT\ Projects\ SAVING PROGRAM SETTINGS

22. Then you can run the analysis: Actions => Calculate => Cross-validated RUNNING CPT

23. CPT begins the specified analysis in a new “Results Window”. Here you can see the steps of the analysis and of the optimization procedure. DATA ANALYSIS

24. Optimizing the number of EOF modes: 1. CPT uses EOF #1 to make cross-validated forecasts then calculates a “goodness index” summarizing how good all the forecasts are (the closer to 1 the better). Then CPT uses EOF #1 and #2 to remake cross-validated forecasts and calculates a new goodness index for these, and so on until using all four EOFs. 2. At each step CPT compares the goodness indices and retains under the column “OPTIMUM” the highest goodness index and the corresponding number of EOFs (in the example above, 1). 3. CPT uses this number of EOFs (i.e. 1) to build the model. DATA ANALYSIS

25. The menu Tools => Graphics => Scree plots displays the percentage of variance associated with each EOF plotted. RESULTS : graphics

26. 1. The menu Tools => Graphics => X EOF loadings and scores displays the loading pattern of each EOF and the temporal series. 2. CPT allows you to customize and save each graphic by: right-clicking on the mouse selecting the graphic to customize / save RESULTS : graphics

27. To change the title of the graph 1. right-click the mouse 2. go to EOF Loadings 3. click on Title CHANGING THE TITLE

28. You can choose the name of the graphic output file by clicking on browse. You can adjust the quality of the JPEG graphic as well. All the output files are saved by default under: C:\CPT\Output\ SAVING GRAPHICS

29. To see the results go to the menu “Tools”: Validation : shows skill, hindcasts and observed series Contingency Tables : shows contingency tables Graphics : shows the EOFs time series, loading patterns and scree plot RESULTS

30. To see the series forecasted and observed at each station/grid go to: Tools => Validation => Cross-Validated => Performance Measures This menu displays some statistics of the forecast, such as correlation coefficient, RMSE, ROC etc (for more details refer to the help page). RESULTS

31. Customise => Graphics => Reverse Colors If you are forecasting temperature instead of precipitation, then it would be more intuitive to have red (hot/above) and blue (cold/below), so you might want to invert the default colors. You might also want black and white images if they are to be included in a report or publication. REVERSING THE COLORS

32. For indications of uncertainty in the performance measures go to: Tools => Validation => Cross-Validated => Bootstrap INDICATIONS OF UNCERTAINTY

33. Customize => Resampling Settings CPT allows you to adjust the bootstrap settings. ADJUSTING THE BOOTSTRAP SETTINGS

34. The menu File => Data Output allows you to save output data: 1.EOFs: time series, loading patterns, variance 2.The parameters (coefficients) of the model (example: Y=Ax+b) 3.The input data (with the missing values filled) 4.Cross-validated forecasted time series RESULTS : data files

35. In order to save the outputs in separate files, you have to specify a file name by clicking on browse. By default CPT saves the output files under: C:\Documents and settings\user\Application Data\CPT\ Output\ SAVING OUTPUT FILES

36. Once your model is built, you can make a forecast using a forecast file with new records of the X variables: File => Open Forecast File FORECAST

37. A new window is opened. By default CPT selects the same input predictor file. You can change it by clicking browse. FORECAST

38. You then select: (a) the starting year of the forecasts (b) the number of years to forecast FORECAST

39. Once the file is selected and the years to forecast are chosen go to the menu Tools => Forecast => Series or Maps. FORECAST

40. The option Series shows the predicted values (cross) for the current station as well as forecast possibilities, confidence limits for the forecast and, in the “Thresholds” box, the “category thresholds” as well as the climatological probabilities for the 3 categories. Below Normal Above Normal Predicted Value FORECAST

41. There are two ways to change how the categories are defined. The first way is to change the climatological probabilities. Customize => Thresholds CHANGING CATEGORY DEFINITIONS CPT recalculates the thresholds

42. The second way is to define the actual thresholds. CHANGING CATEGORY DEFINITIONS CPT recalculates the climatological probabilities.

43. To draw error bars on the forecast, right click on the graph: Customize => Prediction Intervals An error bar is indicated. PREDICTION INTERVALS

44. You can also change the width of the prediction interval. Customize => Forecast Settings The default setting of 68.2% gives standard error bars. CHANGING THE PREDICTION INTERVAL

45. The forecast values map lists the actual forecast values for each category at each location as well as the spatial distribution of the values. FORECAST MAPS

46. To draw the probabilities of exceedance go to: Tools =>Forecast => Exceedances EXCEEDANCE PROBABILITIES

47. For further details, read the help page of each menu and option. Subscribe to the user-list to be advised of updates: http://iri.columbia.edu/outreach/software/index.html We want to hear from you. Your comments and questions help us to improve the CPT so do not hesitate to write to us at: cpt@iri.columbia.edu CONCLUSIONS

49. The forecast can be expressed as anomalies, rather than absolute values: Customize => Forecast Settings => Anomalies EXPRESSING THE FORECAST AS ANOMALIES

50. The thresholds, as well as the forecast ranges, are now defined as anomalies. EXPRESSING THE FORECAST AS ANOMALIES

51. SAVING FORECASTS To save the forecasts, right click, and specify the required output files.

52. CHANGING THE CLIMATOLOGICAL PERIOD By default, the forecast probabilities are calculated relative to a climatological period that is the same as the training period. To change the climatological period go to: Customise => Climatological Period

53. Tools => Forecast => Maps The option Maps lets you see maps of your forecasts – either maps of the probabilities or maps of the actual forecast values. The forecast probabilities map lists the probabilities for each category at each location as well as the spatial distribution of the probabilities. In this example it is evident that in 2000 the below-normal category has the lowest probability over most of north- east Brazil. FORECAST MAPS

54. The forecast values map lists the actual forecast values for each category at each location as well as the spatial distribution of the values. FORECAST MAPS

55. To draw the probabilities of exceedance go to: Tools =>Forecast => Exceedances EXCEEDANCE PROBABILITIES

56. For further details, read the help page of each menu and option. Subscribe to the user-list to be advised of updates: http://iri.columbia.edu/outreach/software/index.html We want to hear from you. Your comments and questions help us to improve the CPT so do not hesitate to write to us at: cpt@iri.columbia.edu CONCLUSIONS